Add ACPI NUMA support for RISC-V

PATCH Superseded Haibo Xu 5 participants 20 messages Jan 31, 2024 – Mar 5, 2024

Subsystems (3)

ACPI RISC-V ACPI component architecture (acpica)

lore

Share & tools

lore

Haibo Xu

author

2024-01-31 02:18 UTC #0 lore raw

cover Add ACPI NUMA support for RISC-V

This patch series enable RISC-V ACPI NUMA support which was based on the recently approved ACPI ECR[1].

Patch 1/4 is generated from the acpica PR[2] and should be merged through the acpica project. Due to this dependency, other 3 patches can only be merged after the corresponding ACPICA patch was pulled into linux.

Patch 2/4 add the common SRAT RINTC affinity structure handler. Patch 3/4 add RISC-V specific acpi_numa.c file to parse NUMA information from SRAT and SLIT ACPI tables.
Patch 4/4 add corresponding ACPI_NUMA config for RISC-V Kconfig.

Based-on: https://github.com/linux-riscv/linux-riscv/tree/for-next

[1] https://mantis.uefi.org/mantis/view.php?id=2433
[2] https://github.com/acpica/acpica/pull/926

Testing:
Since the ACPI AIA/PLIC support patch set is still under upstream review, hence it is tested using the poll based HVC SBI console and RAM disk.
1) Build latest Qemu with the following patch backported
https://lore.kernel.org/all/[email protected]/
https://github.com/vlsunil/qemu/commit/42bd4eeefd5d4410a68f02d54fee406d8a1269b0

2) Build latest EDK-II
https://github.com/tianocore/edk2/blob/master/OvmfPkg/RiscVVirt/README.md

3) Build Linux with the following configs enabled
CONFIG_RISCV_SBI_V01=y
CONFIG_SERIAL_EARLYCON_RISCV_SBI=y
CONFIG_HVC_RISCV_SBI=y

4) Build buildroot rootfs.cpio

5) Launch the Qemu machine
qemu-system-riscv64 -nographic \
-machine virt,pflash0=pflash0,pflash1=pflash1 -smp 4 -m 8G \
-blockdev node-name=pflash0,driver=file,read-only=on,filename=RISCV_VIRT_CODE.fd \
-blockdev node-name=pflash1,driver=file,filename=RISCV_VIRT_VARS.fd \
-object memory-backend-ram,size=4G,id=m0 \
-object memory-backend-ram,size=4G,id=m1 \
-numa node,memdev=m0,cpus=0-1,nodeid=0 \
-numa node,memdev=m1,cpus=2-3,nodeid=1 \
-numa dist,src=0,dst=1,val=30 \
-kernel linux/arch/riscv/boot/Image \
-initrd buildroot/output/images/rootfs.cpio \
-append "root=/dev/ram ro console=hvc0 earlycon=sbi"

[    0.000000] ACPI: SRAT: Node 0 PXM 0 [mem 0x80000000-0x17fffffff]
[    0.000000] ACPI: SRAT: Node 1 PXM 1 [mem 0x180000000-0x27fffffff]
[    0.000000] NUMA: NODE_DATA [mem 0x17fe3bc40-0x17fe3cfff]
[    0.000000] NUMA: NODE_DATA [mem 0x27fff4c40-0x27fff5fff]
...
[    0.000000] ACPI: NUMA: SRAT: PXM 0 -> HARTID 0x0 -> Node 0
[    0.000000] ACPI: NUMA: SRAT: PXM 0 -> HARTID 0x1 -> Node 0
[    0.000000] ACPI: NUMA: SRAT: PXM 1 -> HARTID 0x2 -> Node 1
[    0.000000] ACPI: NUMA: SRAT: PXM 1 -> HARTID 0x3 -> Node 1

Haibo Xu (4):
ACPICA: SRAT: Add RISC-V RINTC affinity structure
ACPI: NUMA: Add handler for SRAT RINTC affinity structure
ACPI: RISCV: Add NUMA support based on SRAT and SLIT
ACPI: RISCV: Enable ACPI based NUMA

 arch/riscv/Kconfig            |   1 +
 arch/riscv/include/asm/acpi.h |  15 +++-
 arch/riscv/kernel/Makefile    |   1 +
 arch/riscv/kernel/acpi.c      |   5 --
 arch/riscv/kernel/acpi_numa.c | 133 ++++++++++++++++++++++++++++++++++
 arch/riscv/kernel/setup.c     |   4 +-
 arch/riscv/kernel/smpboot.c   |   2 -
 drivers/acpi/numa/Kconfig     |   2 +-
 drivers/acpi/numa/srat.c      |  35 ++++++++-
 include/acpi/actbl3.h         |  18 ++++-
 include/linux/acpi.h          |   7 ++
 11 files changed, 209 insertions(+), 14 deletions(-)
 create mode 100644 arch/riscv/kernel/acpi_numa.c

Signature · 1 line

2.34.1

Haibo Xu

author

2024-01-31 02:18 UTC #1 lore raw

PATCH 1/4 ACPICA: SRAT: Add RISC-V RINTC affinity structure

ACPICA commit 93caddbf2f620769052c59ec471f018281dc3a24

Add definition of RISC-V Interrupt Controller(RINTC) affinity structure which was approved by UEFI forum and will be part of next ACPI spec version(6.6).

Link https://github.com/acpica/acpica/commit/93caddbf

Signed-off-by Haibo Xu <[email protected]>

---
 include/acpi/actbl3.h | 18 +++++++++++++++++-
 1 file changed, 17 insertions(+), 1 deletion(-)

include/acpi/actbl3.h +17 -1

index c080d579a546..5202e3fc9b41 100644
@@ -192,7 +192,8 @@ enum acpi_srat_type {
 	ACPI_SRAT_TYPE_GIC_ITS_AFFINITY = 4,	/* ACPI 6.2 */
 	ACPI_SRAT_TYPE_GENERIC_AFFINITY = 5,	/* ACPI 6.3 */
 	ACPI_SRAT_TYPE_GENERIC_PORT_AFFINITY = 6,	/* ACPI 6.4 */
-	ACPI_SRAT_TYPE_RESERVED = 7	/* 7 and greater are reserved */
+	ACPI_SRAT_TYPE_RINTC_AFFINITY = 7,	/* ACPI 6.6 */
+	ACPI_SRAT_TYPE_RESERVED = 8	/* 8 and greater are reserved */
 };
 
 /*

@@ -296,6 +297,21 @@ struct acpi_srat_generic_affinity {
 #define ACPI_SRAT_GENERIC_AFFINITY_ENABLED     (1)	/* 00: Use affinity structure */
 #define ACPI_SRAT_ARCHITECTURAL_TRANSACTIONS   (1<<1)	/* ACPI 6.4 */
 
+/* 7: RINTC Affinity (ACPI 6.6) */
+
+struct acpi_srat_rintc_affinity {
+	struct acpi_subtable_header header;
+	u16 reserved;           /* Reserved, must be zero */
+	u32 proximity_domain;
+	u32 acpi_processor_uid;
+	u32 flags;
+	u32 clock_domain;
+};
+
+/* Flags for struct acpi_srat_rintc_affinity */
+
+#define ACPI_SRAT_RINTC_ENABLED     (1)	/* 00: Use affinity structure */
+
 /*******************************************************************************
  *
  * STAO - Status Override Table (_STA override) - ACPI 6.0

Signature · 1 line

2.34.1

Rafael J. Wysocki

maintainer

2024-02-12 13:31 UTC #2 lore raw

Re: [PATCH 1/4] ACPICA: SRAT: Add RISC-V RINTC affinity structure

Quoted 2 lines · Haibo Xu

ACPICA commit 93caddbf2f620769052c59ec471f018281dc3a24

Not really.

Quoted 5 lines

Add definition of RISC-V Interrupt Controller(RINTC) affinity structure which was approved by UEFI forum and will be part of next ACPI spec version(6.6).
Link https://github.com/acpica/acpica/commit/93caddbf

And this doesn't point to an upstream ACPICA PR.

Quoted 44 lines

Signed-off-by Haibo Xu <[email protected]>

---
 include/acpi/actbl3.h | 18 +++++++++++++++++-
 1 file changed, 17 insertions(+), 1 deletion(-)

include/acpi/actbl3.h +17 -2

index c080d579a546..5202e3fc9b41 100644
@@ -192,7 +192,8 @@ enum acpi_srat_type {
        ACPI_SRAT_TYPE_GIC_ITS_AFFINITY = 4,    /* ACPI 6.2 */
        ACPI_SRAT_TYPE_GENERIC_AFFINITY = 5,    /* ACPI 6.3 */
        ACPI_SRAT_TYPE_GENERIC_PORT_AFFINITY = 6,       /* ACPI 6.4 */
-       ACPI_SRAT_TYPE_RESERVED = 7     /* 7 and greater are reserved */
+       ACPI_SRAT_TYPE_RINTC_AFFINITY = 7,      /* ACPI 6.6 */
+       ACPI_SRAT_TYPE_RESERVED = 8     /* 8 and greater are reserved */
 };
 
 /*

@@ -296,6 +297,21 @@ struct acpi_srat_generic_affinity {
 #define ACPI_SRAT_GENERIC_AFFINITY_ENABLED     (1)     /* 00: Use affinity structure */
 #define ACPI_SRAT_ARCHITECTURAL_TRANSACTIONS   (1<<1)  /* ACPI 6.4 */
 
+/* 7: RINTC Affinity (ACPI 6.6) */
+
+struct acpi_srat_rintc_affinity {
+       struct acpi_subtable_header header;
+       u16 reserved;           /* Reserved, must be zero */
+       u32 proximity_domain;
+       u32 acpi_processor_uid;
+       u32 flags;
+       u32 clock_domain;
+};
+
+/* Flags for struct acpi_srat_rintc_affinity */
+
+#define ACPI_SRAT_RINTC_ENABLED     (1)        /* 00: Use affinity structure */
+
 /*******************************************************************************
  *
  * STAO - Status Override Table (_STA override) - ACPI 6.0
--

2.34.1

Haibo Xu

2024-02-18 07:18 UTC #3 lore raw

Re: [PATCH 1/4] ACPICA: SRAT: Add RISC-V RINTC affinity structure

Quoted 7 lines · Rafael J. Wysocki

Quoted 2 lines · Haibo Xu
ACPICA commit 93caddbf2f620769052c59ec471f018281dc3a24
Not really.

Quoted 8 lines

Quoted 5 lines
Add definition of RISC-V Interrupt Controller(RINTC) affinity structure which was approved by UEFI forum and will be part of next ACPI spec version(6.6).
Link https://github.com/acpica/acpica/commit/93caddbf
And this doesn't point to an upstream ACPICA PR.

Hi Rafael,

This patch was generated by the acpica/generate/linux/gen-patch.sh and should be merged
through the ACPICA project. To include it in this series is just to serve as a reference.

BTW, the corresponding ACPICA patch is still under review and please refer to the following
link for the status.

https://github.com/acpica/acpica/pull/926

Regards,
Haibo

Quoted 44 lines

Signed-off-by Haibo Xu <[email protected]>

---
 include/acpi/actbl3.h | 18 +++++++++++++++++-
 1 file changed, 17 insertions(+), 1 deletion(-)

include/acpi/actbl3.h +17 -2

index c080d579a546..5202e3fc9b41 100644
@@ -192,7 +192,8 @@ enum acpi_srat_type {
        ACPI_SRAT_TYPE_GIC_ITS_AFFINITY = 4,    /* ACPI 6.2 */
        ACPI_SRAT_TYPE_GENERIC_AFFINITY = 5,    /* ACPI 6.3 */
        ACPI_SRAT_TYPE_GENERIC_PORT_AFFINITY = 6,       /* ACPI 6.4 */
-       ACPI_SRAT_TYPE_RESERVED = 7     /* 7 and greater are reserved */
+       ACPI_SRAT_TYPE_RINTC_AFFINITY = 7,      /* ACPI 6.6 */
+       ACPI_SRAT_TYPE_RESERVED = 8     /* 8 and greater are reserved */
 };
 
 /*

@@ -296,6 +297,21 @@ struct acpi_srat_generic_affinity {
 #define ACPI_SRAT_GENERIC_AFFINITY_ENABLED     (1)     /* 00: Use affinity structure */
 #define ACPI_SRAT_ARCHITECTURAL_TRANSACTIONS   (1<<1)  /* ACPI 6.4 */
 
+/* 7: RINTC Affinity (ACPI 6.6) */
+
+struct acpi_srat_rintc_affinity {
+       struct acpi_subtable_header header;
+       u16 reserved;           /* Reserved, must be zero */
+       u32 proximity_domain;
+       u32 acpi_processor_uid;
+       u32 flags;
+       u32 clock_domain;
+};
+
+/* Flags for struct acpi_srat_rintc_affinity */
+
+#define ACPI_SRAT_RINTC_ENABLED     (1)        /* 00: Use affinity structure */
+
 /*******************************************************************************
  *
  * STAO - Status Override Table (_STA override) - ACPI 6.0
--

2.34.1

Haibo Xu

author

2024-01-31 02:19 UTC #4 lore raw

PATCH 2/4 ACPI: NUMA: Add handler for SRAT RINTC affinity structure

In reply to Haibo Xu

Add RINTC affinity structure handler during parsing SRAT table. The ARCH specific implementation will be added in next patch.

Signed-off-by Haibo Xu <[email protected]>

---
 drivers/acpi/numa/srat.c | 32 +++++++++++++++++++++++++++++++-
 include/linux/acpi.h     |  3 +++
 2 files changed, 34 insertions(+), 1 deletion(-)

drivers/acpi/numa/srat.c +31 -1

index 0214518fc582..503abcf6125d 100644
@@ -165,6 +165,19 @@ acpi_table_print_srat_entry(struct acpi_subtable_header *header)
 		}
 	}
 	break;
+
+	case ACPI_SRAT_TYPE_RINTC_AFFINITY:
+		{
+			struct acpi_srat_rintc_affinity *p =
+			    (struct acpi_srat_rintc_affinity *)header;
+			pr_debug("SRAT Processor (acpi id[0x%04x]) in proximity domain %d %s\n",
+				 p->acpi_processor_uid,
+				 p->proximity_domain,
+				 (p->flags & ACPI_SRAT_RINTC_ENABLED) ?
+				 "enabled" : "disabled");
+		}
+		break;
+
 	default:
 		pr_warn("Found unsupported SRAT entry (type = 0x%x)\n",
 			header->type);

@@ -448,6 +461,21 @@ acpi_parse_gi_affinity(union acpi_subtable_headers *header,
 }
 #endif /* defined(CONFIG_X86) || defined (CONFIG_ARM64) */
 
+static int __init
+acpi_parse_rintc_affinity(union acpi_subtable_headers *header,
+			 const unsigned long end)
+{
+	struct acpi_srat_rintc_affinity *rintc_affinity;
+
+	rintc_affinity = (struct acpi_srat_rintc_affinity *)header;
+	acpi_table_print_srat_entry(&header->common);
+
+	/* let architecture-dependent part to do it */
+	acpi_numa_rintc_affinity_init(rintc_affinity);
+
+	return 0;
+}
+
 static int __initdata parsed_numa_memblks;
 
 static int __init

@@ -501,7 +529,7 @@ int __init acpi_numa_init(void)
 
 	/* SRAT: System Resource Affinity Table */
 	if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
-		struct acpi_subtable_proc srat_proc[4];
+		struct acpi_subtable_proc srat_proc[5];
 
 		memset(srat_proc, 0, sizeof(srat_proc));
 		srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY;

@@ -512,6 +540,8 @@ int __init acpi_numa_init(void)
 		srat_proc[2].handler = acpi_parse_gicc_affinity;
 		srat_proc[3].id = ACPI_SRAT_TYPE_GENERIC_AFFINITY;
 		srat_proc[3].handler = acpi_parse_gi_affinity;
+		srat_proc[4].id = ACPI_SRAT_TYPE_RINTC_AFFINITY;
+		srat_proc[4].handler = acpi_parse_rintc_affinity;
 
 		acpi_table_parse_entries_array(ACPI_SIG_SRAT,
 					sizeof(struct acpi_table_srat),

include/linux/acpi.h +3 -0

index b7165e52b3c6..a65273db55c6 100644
@@ -269,6 +269,9 @@ acpi_numa_gicc_affinity_init(struct acpi_srat_gicc_affinity *pa) { }
 
 int acpi_numa_memory_affinity_init (struct acpi_srat_mem_affinity *ma);
 
+static inline void
+acpi_numa_rintc_affinity_init(struct acpi_srat_rintc_affinity *pa) { }
+
 #ifndef PHYS_CPUID_INVALID
 typedef u32 phys_cpuid_t;
 #define PHYS_CPUID_INVALID (phys_cpuid_t)(-1)

Signature · 1 line

2.34.1

Sunil V L

2024-03-05 04:42 UTC #5 lore raw

Re: [PATCH 2/4] ACPI: NUMA: Add handler for SRAT RINTC affinity structure

Hi Haibo,

Quoted 34 lines · Haibo Xu

Add RINTC affinity structure handler during parsing SRAT table. The ARCH specific implementation will be added in next patch.

Signed-off-by Haibo Xu <[email protected]>

---
 drivers/acpi/numa/srat.c | 32 +++++++++++++++++++++++++++++++-
 include/linux/acpi.h     |  3 +++
 2 files changed, 34 insertions(+), 1 deletion(-)

drivers/acpi/numa/srat.c +13 -0

index 0214518fc582..503abcf6125d 100644
@@ -165,6 +165,19 @@ acpi_table_print_srat_entry(struct acpi_subtable_header *header)
 		}
 	}
 	break;
+
+	case ACPI_SRAT_TYPE_RINTC_AFFINITY:
+		{
+			struct acpi_srat_rintc_affinity *p =
+			    (struct acpi_srat_rintc_affinity *)header;
+			pr_debug("SRAT Processor (acpi id[0x%04x]) in proximity domain %d %s\n",
+				 p->acpi_processor_uid,
+				 p->proximity_domain,
+				 (p->flags & ACPI_SRAT_RINTC_ENABLED) ?
+				 "enabled" : "disabled");
+		}
+		break;
+
 	default:
 		pr_warn("Found unsupported SRAT entry (type = 0x%x)\n",
 			header->type);

drivers/acpi/numa/srat.c acpi_parse_gi_affinity(union acpi_subtable_headers *header, · line 466

Quoted 6 lines

drivers/acpi/numa/srat.c +3 -0

@@ -448,6 +461,21 @@ acpi_parse_gi_affinity(union acpi_subtable_headers *header,
 }
 #endif /* defined(CONFIG_X86) || defined (CONFIG_ARM64) */
 
+static int __init
+acpi_parse_rintc_affinity(union acpi_subtable_headers *header,
+			 const unsigned long end)

Alignment doesn't look right. Could you please run checkpatch on all the patches?

drivers/acpi/numa/srat.c acpi_parse_gi_affinity(union acpi_subtable_headers *header, · line 475

Quoted 9 lines

drivers/acpi/numa/srat.c +9 -0

+{
+	struct acpi_srat_rintc_affinity *rintc_affinity;
+
+	rintc_affinity = (struct acpi_srat_rintc_affinity *)header;
+	acpi_table_print_srat_entry(&header->common);
+
+	/* let architecture-dependent part to do it */
+	acpi_numa_rintc_affinity_init(rintc_affinity);
+

Is it required to have this commit first prior to architecture functionality? I am wondering whether it is logically better to implement the function first and then consume in next commit?

Quoted 29 lines

drivers/acpi/numa/srat.c +6 -1

+	return 0;
+}
+
 static int __initdata parsed_numa_memblks;
 
 static int __init

@@ -501,7 +529,7 @@ int __init acpi_numa_init(void)
 
 	/* SRAT: System Resource Affinity Table */
 	if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
-		struct acpi_subtable_proc srat_proc[4];
+		struct acpi_subtable_proc srat_proc[5];
 
 		memset(srat_proc, 0, sizeof(srat_proc));
 		srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY;

@@ -512,6 +540,8 @@ int __init acpi_numa_init(void)
 		srat_proc[2].handler = acpi_parse_gicc_affinity;
 		srat_proc[3].id = ACPI_SRAT_TYPE_GENERIC_AFFINITY;
 		srat_proc[3].handler = acpi_parse_gi_affinity;
+		srat_proc[4].id = ACPI_SRAT_TYPE_RINTC_AFFINITY;
+		srat_proc[4].handler = acpi_parse_rintc_affinity;
 
 		acpi_table_parse_entries_array(ACPI_SIG_SRAT,
 					sizeof(struct acpi_table_srat),

include/linux/acpi.h +0 -0

index b7165e52b3c6..a65273db55c6 100644

include/linux/acpi.h acpi_numa_gicc_affinity_init(struct acpi_srat_gicc_affinity *pa) { } · line 274

Quoted 6 lines

include/linux/acpi.h +3 -0

@@ -269,6 +269,9 @@ acpi_numa_gicc_affinity_init(struct acpi_srat_gicc_affinity *pa) { }
 
 int acpi_numa_memory_affinity_init (struct acpi_srat_mem_affinity *ma);
 
+static inline void
+acpi_numa_rintc_affinity_init(struct acpi_srat_rintc_affinity *pa) { }
+

I think this can be fit in single like as we can have upto 100 characters.

Quoted 6 lines

include/linux/acpi.h +0 -0

 #ifndef PHYS_CPUID_INVALID
 typedef u32 phys_cpuid_t;
 #define PHYS_CPUID_INVALID (phys_cpuid_t)(-1)

Signature · 1 line

2.34.1

Haibo Xu

2024-03-05 08:42 UTC #6 lore raw

Re: [PATCH 2/4] ACPI: NUMA: Add handler for SRAT RINTC affinity structure

Quoted 41 lines · Sunil V L

Hi Haibo,

Quoted 37 lines · Haibo Xu

Add RINTC affinity structure handler during parsing SRAT table. The ARCH specific implementation will be added in next patch.

Signed-off-by Haibo Xu <[email protected]>

---
 drivers/acpi/numa/srat.c | 32 +++++++++++++++++++++++++++++++-
 include/linux/acpi.h     |  3 +++
 2 files changed, 34 insertions(+), 1 deletion(-)

drivers/acpi/numa/srat.c +13 -0

index 0214518fc582..503abcf6125d 100644
@@ -165,6 +165,19 @@ acpi_table_print_srat_entry(struct acpi_subtable_header *header)
              }
      }
      break;
+
+     case ACPI_SRAT_TYPE_RINTC_AFFINITY:
+             {
+                     struct acpi_srat_rintc_affinity *p =
+                         (struct acpi_srat_rintc_affinity *)header;
+                     pr_debug("SRAT Processor (acpi id[0x%04x]) in proximity domain %d %s\n",
+                              p->acpi_processor_uid,
+                              p->proximity_domain,
+                              (p->flags & ACPI_SRAT_RINTC_ENABLED) ?
+                              "enabled" : "disabled");
+             }
+             break;
+
      default:
              pr_warn("Found unsupported SRAT entry (type = 0x%x)\n",
                      header->type);

@@ -448,6 +461,21 @@ acpi_parse_gi_affinity(union acpi_subtable_headers *header,
 }
 #endif /* defined(CONFIG_X86) || defined (CONFIG_ARM64) */

Quoted 6 lines

drivers/acpi/numa/srat.c acpi_parse_gi_affinity(union acpi_subtable_headers *header, · line 466
Quoted 3 lines
drivers/acpi/numa/srat.c +3 -0
+static int __init
+acpi_parse_rintc_affinity(union acpi_subtable_headers *header,
+                      const unsigned long end)
Alignment doesn't look right. Could you please run checkpatch on all the patches?

Seems something is wrong with my vim configuration. Will fix it in v2.

Quoted 4 lines

drivers/acpi/numa/srat.c +4 -0

+{
+     struct acpi_srat_rintc_affinity *rintc_affinity;
+
+     rintc_affinity = (struct acpi_srat_rintc_affinity *)header;

Quoted 9 lines

drivers/acpi/numa/srat.c acpi_parse_gi_affinity(union acpi_subtable_headers *header, · line 475
Quoted 5 lines
drivers/acpi/numa/srat.c +5 -0
+     acpi_table_print_srat_entry(&header->common);
+
+     /* let architecture-dependent part to do it */
+     acpi_numa_rintc_affinity_init(rintc_affinity);
+
Is it required to have this commit first prior to architecture functionality? I am wondering whether it is logically better to implement the function first and then consume in next commit?

No dependency between this commit and the next commit. Will change the order in v2.

Quoted 30 lines

drivers/acpi/numa/srat.c +6 -1

+     return 0;
+}
+
 static int __initdata parsed_numa_memblks;
 
 static int __init

@@ -501,7 +529,7 @@ int __init acpi_numa_init(void)
 
      /* SRAT: System Resource Affinity Table */
      if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
-             struct acpi_subtable_proc srat_proc[4];
+             struct acpi_subtable_proc srat_proc[5];
 
              memset(srat_proc, 0, sizeof(srat_proc));
              srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY;

@@ -512,6 +540,8 @@ int __init acpi_numa_init(void)
              srat_proc[2].handler = acpi_parse_gicc_affinity;
              srat_proc[3].id = ACPI_SRAT_TYPE_GENERIC_AFFINITY;
              srat_proc[3].handler = acpi_parse_gi_affinity;
+             srat_proc[4].id = ACPI_SRAT_TYPE_RINTC_AFFINITY;
+             srat_proc[4].handler = acpi_parse_rintc_affinity;
 
              acpi_table_parse_entries_array(ACPI_SIG_SRAT,
                                      sizeof(struct acpi_table_srat),

include/linux/acpi.h +0 -0

index b7165e52b3c6..a65273db55c6 100644
@@ -269,6 +269,9 @@ acpi_numa_gicc_affinity_init(struct acpi_srat_gicc_affinity *pa) { }

Quoted 8 lines

include/linux/acpi.h acpi_numa_gicc_affinity_init(struct acpi_srat_gicc_affinity *pa) { } · line 274
Quoted 5 lines
include/linux/acpi.h +3 -0
 int acpi_numa_memory_affinity_init (struct acpi_srat_mem_affinity *ma);
 
+static inline void
+acpi_numa_rintc_affinity_init(struct acpi_srat_rintc_affinity *pa) { }
+
I think this can be fit in single like as we can have upto 100 characters.

Sure. will fix it in v2.

Thanks,
Haibo

Quoted 6 lines

include/linux/acpi.h +0 -1

 #ifndef PHYS_CPUID_INVALID
 typedef u32 phys_cpuid_t;
 #define PHYS_CPUID_INVALID (phys_cpuid_t)(-1)
--

2.34.1

Haibo Xu

author

2024-01-31 02:19 UTC #7 lore raw

PATCH 3/4 ACPI: RISCV: Add NUMA support based on SRAT and SLIT

In reply to Haibo Xu

Add acpi_numa.c file to enable parse NUMA information from ACPI SRAT and SLIT tables. SRAT table provide CPUs(Hart) and memory nodes to proximity domain mapping, while SLIT table provide the distance metrics between proximity domains.

Signed-off-by Haibo Xu <[email protected]>

---
 arch/riscv/include/asm/acpi.h |  15 +++-
 arch/riscv/kernel/Makefile    |   1 +
 arch/riscv/kernel/acpi.c      |   5 --
 arch/riscv/kernel/acpi_numa.c | 133 ++++++++++++++++++++++++++++++++++
 arch/riscv/kernel/setup.c     |   4 +-
 arch/riscv/kernel/smpboot.c   |   2 -
 drivers/acpi/numa/srat.c      |   3 +-
 include/linux/acpi.h          |   4 +
 8 files changed, 156 insertions(+), 11 deletions(-)
 create mode 100644 arch/riscv/kernel/acpi_numa.c

arch/riscv/include/asm/acpi.h +13 -2

index 7dad0cf9d701..e0a1f84404f3 100644
@@ -61,11 +61,14 @@ static inline void arch_fix_phys_package_id(int num, u32 slot) { }
 
 void acpi_init_rintc_map(void);
 struct acpi_madt_rintc *acpi_cpu_get_madt_rintc(int cpu);
-u32 get_acpi_id_for_cpu(int cpu);
+static inline u32 get_acpi_id_for_cpu(int cpu)
+{
+	return acpi_cpu_get_madt_rintc(cpu)->uid;
+}
+
 int acpi_get_riscv_isa(struct acpi_table_header *table,
 		       unsigned int cpu, const char **isa);
 
-static inline int acpi_numa_get_nid(unsigned int cpu) { return NUMA_NO_NODE; }
 void acpi_get_cbo_block_size(struct acpi_table_header *table, u32 *cbom_size,
 			     u32 *cboz_size, u32 *cbop_size);
 #else

@@ -87,4 +90,12 @@ static inline void acpi_get_cbo_block_size(struct acpi_table_header *table,
 
 #endif /* CONFIG_ACPI */
 
+#ifdef CONFIG_ACPI_NUMA
+int acpi_numa_get_nid(unsigned int cpu);
+void acpi_map_cpus_to_nodes(void);
+#else
+static inline int acpi_numa_get_nid(unsigned int cpu) { return NUMA_NO_NODE; }
+static inline void acpi_map_cpus_to_nodes(void) { }
+#endif /* CONFIG_ACPI_NUMA */
+
 #endif /*_ASM_ACPI_H*/

arch/riscv/kernel/Makefile +1 -0

index f71910718053..5d3e9cf89b76 100644
@@ -105,3 +105,4 @@ obj-$(CONFIG_COMPAT)		+= compat_vdso/
 
 obj-$(CONFIG_64BIT)		+= pi/
 obj-$(CONFIG_ACPI)		+= acpi.o
+obj-$(CONFIG_ACPI_NUMA)	+= acpi_numa.o

arch/riscv/kernel/acpi.c +0 -5

index e619edc8b0cc..040bdbfea2b4 100644
@@ -191,11 +191,6 @@ struct acpi_madt_rintc *acpi_cpu_get_madt_rintc(int cpu)
 	return &cpu_madt_rintc[cpu];
 }
 
-u32 get_acpi_id_for_cpu(int cpu)
-{
-	return acpi_cpu_get_madt_rintc(cpu)->uid;
-}
-
 /*
  * __acpi_map_table() will be called before paging_init(), so early_ioremap()
  * or early_memremap() should be called here to for ACPI table mapping.

arch/riscv/kernel/acpi_numa.c new file +133 -0

133 lines not on this page · raw message

arch/riscv/kernel/setup.c +3 -1

index 4f73c0ae44b2..a2cde65b69e9 100644
@@ -281,8 +281,10 @@ void __init setup_arch(char **cmdline_p)
 	setup_smp();
 #endif
 
-	if (!acpi_disabled)
+	if (!acpi_disabled) {
 		acpi_init_rintc_map();
+		acpi_map_cpus_to_nodes();
+	}
 
 	riscv_init_cbo_blocksizes();
 	riscv_fill_hwcap();

arch/riscv/kernel/smpboot.c +0 -2

index 519b6bd946e5..b188d83d1ec4 100644
@@ -101,7 +101,6 @@ static int __init acpi_parse_rintc(union acpi_subtable_headers *header, const un
 	if (hart == cpuid_to_hartid_map(0)) {
 		BUG_ON(found_boot_cpu);
 		found_boot_cpu = true;
-		early_map_cpu_to_node(0, acpi_numa_get_nid(cpu_count));
 		return 0;
 	}

@@ -111,7 +110,6 @@ static int __init acpi_parse_rintc(union acpi_subtable_headers *header, const un
 	}
 
 	cpuid_to_hartid_map(cpu_count) = hart;
-	early_map_cpu_to_node(cpu_count, acpi_numa_get_nid(cpu_count));
 	cpu_count++;
 
 	return 0;

drivers/acpi/numa/srat.c +2 -1

index 503abcf6125d..1f0462cef47c 100644
@@ -219,7 +219,8 @@ int __init srat_disabled(void)
 	return acpi_numa < 0;
 }
 
-#if defined(CONFIG_X86) || defined(CONFIG_ARM64) || defined(CONFIG_LOONGARCH)
+#if defined(CONFIG_X86) || defined(CONFIG_ARM64) || defined(CONFIG_LOONGARCH) || \
+	defined(CONFIG_RISCV)
 /*
  * Callback for SLIT parsing.  pxm_to_node() returns NUMA_NO_NODE for
  * I/O localities since SRAT does not list them.  I/O localities are

include/linux/acpi.h +4 -0

index a65273db55c6..be78a9d28927 100644
@@ -269,8 +269,12 @@ acpi_numa_gicc_affinity_init(struct acpi_srat_gicc_affinity *pa) { }
 
 int acpi_numa_memory_affinity_init (struct acpi_srat_mem_affinity *ma);
 
+#ifdef CONFIG_RISCV
+void acpi_numa_rintc_affinity_init(struct acpi_srat_rintc_affinity *pa);
+#else
 static inline void
 acpi_numa_rintc_affinity_init(struct acpi_srat_rintc_affinity *pa) { }
+#endif
 
 #ifndef PHYS_CPUID_INVALID
 typedef u32 phys_cpuid_t;

Signature · 1 line

2.34.1

Sunil V L

2024-03-05 05:24 UTC #8 lore raw

Re: [PATCH 3/4] ACPI: RISCV: Add NUMA support based on SRAT and SLIT

Quoted 123 lines · Haibo Xu

Signed-off-by Haibo Xu <[email protected]>

---
 arch/riscv/include/asm/acpi.h |  15 +++-
 arch/riscv/kernel/Makefile    |   1 +
 arch/riscv/kernel/acpi.c      |   5 --
 arch/riscv/kernel/acpi_numa.c | 133 ++++++++++++++++++++++++++++++++++
 arch/riscv/kernel/setup.c     |   4 +-
 arch/riscv/kernel/smpboot.c   |   2 -
 drivers/acpi/numa/srat.c      |   3 +-
 include/linux/acpi.h          |   4 +
 8 files changed, 156 insertions(+), 11 deletions(-)
 create mode 100644 arch/riscv/kernel/acpi_numa.c

arch/riscv/include/asm/acpi.h +13 -2

index 7dad0cf9d701..e0a1f84404f3 100644
@@ -61,11 +61,14 @@ static inline void arch_fix_phys_package_id(int num, u32 slot) { }
 
 void acpi_init_rintc_map(void);
 struct acpi_madt_rintc *acpi_cpu_get_madt_rintc(int cpu);
-u32 get_acpi_id_for_cpu(int cpu);
+static inline u32 get_acpi_id_for_cpu(int cpu)
+{
+	return acpi_cpu_get_madt_rintc(cpu)->uid;
+}
+
 int acpi_get_riscv_isa(struct acpi_table_header *table,
 		       unsigned int cpu, const char **isa);
 
-static inline int acpi_numa_get_nid(unsigned int cpu) { return NUMA_NO_NODE; }
 void acpi_get_cbo_block_size(struct acpi_table_header *table, u32 *cbom_size,
 			     u32 *cboz_size, u32 *cbop_size);
 #else

@@ -87,4 +90,12 @@ static inline void acpi_get_cbo_block_size(struct acpi_table_header *table,
 
 #endif /* CONFIG_ACPI */
 
+#ifdef CONFIG_ACPI_NUMA
+int acpi_numa_get_nid(unsigned int cpu);
+void acpi_map_cpus_to_nodes(void);
+#else
+static inline int acpi_numa_get_nid(unsigned int cpu) { return NUMA_NO_NODE; }
+static inline void acpi_map_cpus_to_nodes(void) { }
+#endif /* CONFIG_ACPI_NUMA */
+
 #endif /*_ASM_ACPI_H*/

arch/riscv/kernel/Makefile +1 -0

index f71910718053..5d3e9cf89b76 100644
@@ -105,3 +105,4 @@ obj-$(CONFIG_COMPAT)		+= compat_vdso/
 
 obj-$(CONFIG_64BIT)		+= pi/
 obj-$(CONFIG_ACPI)		+= acpi.o
+obj-$(CONFIG_ACPI_NUMA)	+= acpi_numa.o

arch/riscv/kernel/acpi.c +0 -5

index e619edc8b0cc..040bdbfea2b4 100644
@@ -191,11 +191,6 @@ struct acpi_madt_rintc *acpi_cpu_get_madt_rintc(int cpu)
 	return &cpu_madt_rintc[cpu];
 }
 
-u32 get_acpi_id_for_cpu(int cpu)
-{
-	return acpi_cpu_get_madt_rintc(cpu)->uid;
-}
-
 /*
  * __acpi_map_table() will be called before paging_init(), so early_ioremap()
  * or early_memremap() should be called here to for ACPI table mapping.

arch/riscv/kernel/acpi_numa.c new file +40 -0

new file mode 100644
index 000000000000..493642a61457
@@ -0,0 +1,133 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * ACPI 6.6 based NUMA setup for RISCV
+ * Lots of code was borrowed from arch/arm64/kernel/acpi_numa.c
+ *
+ * Copyright 2004 Andi Kleen, SuSE Labs.
+ * Copyright (C) 2013-2016, Linaro Ltd.
+ *		Author: Hanjun Guo <[email protected]>
+ * Copyright (C) 2024 Intel Corporation.
+ *
+ * Reads the ACPI SRAT table to figure out what memory belongs to which CPUs.
+ *
+ * Called from acpi_numa_init while reading the SRAT and SLIT tables.
+ * Assumes all memory regions belonging to a single proximity domain
+ * are in one chunk. Holes between them will be included in the node.
+ */
+
+#define pr_fmt(fmt) "ACPI: NUMA: " fmt
+
+#include <linux/acpi.h>
+#include <linux/bitmap.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/memblock.h>
+#include <linux/mmzone.h>
+#include <linux/module.h>
+#include <linux/topology.h>
+
+#include <asm/numa.h>
+
+static int acpi_early_node_map[NR_CPUS] __initdata = { NUMA_NO_NODE };
+
+int __init acpi_numa_get_nid(unsigned int cpu)
+{
+	return acpi_early_node_map[cpu];
+}
+
+static inline int get_cpu_for_acpi_id(u32 uid)
+{
+	int cpu;

arch/riscv/kernel/acpi_numa.c line 50

Quoted 10 lines

arch/riscv/kernel/acpi_numa.c new file +10 -0

+
+	for (cpu = 0; cpu < nr_cpu_ids; cpu++)
+		if (uid == get_acpi_id_for_cpu(cpu))
+			return cpu;
+
+	return -EINVAL;
+}
+
+static int __init acpi_parse_rintc_pxm(union acpi_subtable_headers *header,
+				      const unsigned long end)

Please check alignment.

Quoted 40 lines

arch/riscv/kernel/acpi_numa.c new file +40 -0

+{
+	struct acpi_srat_rintc_affinity *pa;
+	int cpu, pxm, node;
+
+	if (srat_disabled())
+		return -EINVAL;
+
+	pa = (struct acpi_srat_rintc_affinity *)header;
+	if (!pa)
+		return -EINVAL;
+
+	if (!(pa->flags & ACPI_SRAT_RINTC_ENABLED))
+		return 0;
+
+	pxm = pa->proximity_domain;
+	node = pxm_to_node(pxm);
+
+	/*
+	 * If we can't map the UID to a logical cpu this
+	 * means that the UID is not part of possible cpus
+	 * so we do not need a NUMA mapping for it, skip
+	 * the SRAT entry and keep parsing.
+	 */
+	cpu = get_cpu_for_acpi_id(pa->acpi_processor_uid);
+	if (cpu < 0)
+		return 0;
+
+	acpi_early_node_map[cpu] = node;
+	pr_info("SRAT: PXM %d -> HARTID 0x%lx -> Node %d\n", pxm,
+		cpuid_to_hartid_map(cpu), node);
+
+	return 0;
+}
+
+void __init acpi_map_cpus_to_nodes(void)
+{
+	int i;
+
+	/*
+	 * In ACPI, SMP and CPU NUMA information is provided in separate

arch/riscv/kernel/acpi_numa.c line 100

Quoted 10 lines

arch/riscv/kernel/acpi_numa.c new file +10 -0

+	 * static tables, namely the MADT and the SRAT.
+	 *
+	 * Thus, it is simpler to first create the cpu logical map through
+	 * an MADT walk and then map the logical cpus to their node ids
+	 * as separate steps.
+	 */
+	acpi_table_parse_entries(ACPI_SIG_SRAT, sizeof(struct acpi_table_srat),
+					    ACPI_SRAT_TYPE_RINTC_AFFINITY,
+					    acpi_parse_rintc_pxm, 0);
+

Alignment here as well.

Quoted 5 lines

arch/riscv/kernel/acpi_numa.c new file +5 -0

+	for (i = 0; i < nr_cpu_ids; i++)
+		early_map_cpu_to_node(i, acpi_numa_get_nid(i));
+}
+
+/* Callback for Proximity Domain -> logical node ID mapping */

arch/riscv/kernel/acpi_numa.c line 115

Quoted 10 lines

arch/riscv/kernel/acpi_numa.c new file +10 -0

+void __init acpi_numa_rintc_affinity_init(struct acpi_srat_rintc_affinity *pa)
+{
+	int pxm, node;
+
+	if (srat_disabled())
+		return;
+
+	if (pa->header.length < sizeof(struct acpi_srat_rintc_affinity)) {
+		pr_err("SRAT: Invalid SRAT header length: %d\n",
+			pa->header.length);

Can we merge these into single line?

Quoted 23 lines

arch/riscv/kernel/acpi_numa.c new file +18 -0

+		bad_srat();
+		return;
+	}
+
+	if (!(pa->flags & ACPI_SRAT_RINTC_ENABLED))
+		return;
+
+	pxm = pa->proximity_domain;
+	node = acpi_map_pxm_to_node(pxm);
+
+	if (node == NUMA_NO_NODE) {
+		pr_err("SRAT: Too many proximity domains %d\n", pxm);
+		bad_srat();
+		return;
+	}
+
+	node_set(node, numa_nodes_parsed);
+}

arch/riscv/kernel/setup.c +0 -0

index 4f73c0ae44b2..a2cde65b69e9 100644

arch/riscv/kernel/setup.c void __init setup_arch(char **cmdline_p) · line 286

Quoted 7 lines

arch/riscv/kernel/setup.c +2 -1

@@ -281,8 +281,10 @@ void __init setup_arch(char **cmdline_p)
 	setup_smp();
 #endif
 
-	if (!acpi_disabled)
+	if (!acpi_disabled) {
 		acpi_init_rintc_map();
+		acpi_map_cpus_to_nodes();

Is it not possible to fill up both in single parsing of MADT?

Quoted 29 lines

arch/riscv/kernel/setup.c +1 -0

+	}
 
 	riscv_init_cbo_blocksizes();
 	riscv_fill_hwcap();

arch/riscv/kernel/smpboot.c +0 -2

index 519b6bd946e5..b188d83d1ec4 100644
@@ -101,7 +101,6 @@ static int __init acpi_parse_rintc(union acpi_subtable_headers *header, const un
 	if (hart == cpuid_to_hartid_map(0)) {
 		BUG_ON(found_boot_cpu);
 		found_boot_cpu = true;
-		early_map_cpu_to_node(0, acpi_numa_get_nid(cpu_count));
 		return 0;
 	}

@@ -111,7 +110,6 @@ static int __init acpi_parse_rintc(union acpi_subtable_headers *header, const un
 	}
 
 	cpuid_to_hartid_map(cpu_count) = hart;
-	early_map_cpu_to_node(cpu_count, acpi_numa_get_nid(cpu_count));
 	cpu_count++;
 
 	return 0;

drivers/acpi/numa/srat.c +0 -0

index 503abcf6125d..1f0462cef47c 100644

drivers/acpi/numa/srat.c int __init srat_disabled(void) · line 223

Quoted 6 lines

drivers/acpi/numa/srat.c +2 -1

@@ -219,7 +219,8 @@ int __init srat_disabled(void)
 	return acpi_numa < 0;
 }
 
-#if defined(CONFIG_X86) || defined(CONFIG_ARM64) || defined(CONFIG_LOONGARCH)
+#if defined(CONFIG_X86) || defined(CONFIG_ARM64) || defined(CONFIG_LOONGARCH) || \
+	defined(CONFIG_RISCV)

Please check alignment. Or make it single line if fits in 100 chars. Also, it looks it covers most of the architectures now. Is it possible to simplify / remove the condition? I hope IA64 is removed now?

May be you need to update the comment at #endif too.

Thanks
Sunil

Quoted 23 lines

drivers/acpi/numa/srat.c +0 -0

 /*
  * Callback for SLIT parsing.  pxm_to_node() returns NUMA_NO_NODE for
  * I/O localities since SRAT does not list them.  I/O localities are

include/linux/acpi.h +4 -0

index a65273db55c6..be78a9d28927 100644
@@ -269,8 +269,12 @@ acpi_numa_gicc_affinity_init(struct acpi_srat_gicc_affinity *pa) { }
 
 int acpi_numa_memory_affinity_init (struct acpi_srat_mem_affinity *ma);
 
+#ifdef CONFIG_RISCV
+void acpi_numa_rintc_affinity_init(struct acpi_srat_rintc_affinity *pa);
+#else
 static inline void
 acpi_numa_rintc_affinity_init(struct acpi_srat_rintc_affinity *pa) { }
+#endif
 
 #ifndef PHYS_CPUID_INVALID
 typedef u32 phys_cpuid_t;

Signature · 1 line

2.34.1

Haibo Xu

2024-03-05 09:54 UTC #9 lore raw

Re: [PATCH 3/4] ACPI: RISCV: Add NUMA support based on SRAT and SLIT

Quoted 129 lines · Sunil V L

Quoted 127 lines · Haibo Xu

Signed-off-by Haibo Xu <[email protected]>

---
 arch/riscv/include/asm/acpi.h |  15 +++-
 arch/riscv/kernel/Makefile    |   1 +
 arch/riscv/kernel/acpi.c      |   5 --
 arch/riscv/kernel/acpi_numa.c | 133 ++++++++++++++++++++++++++++++++++
 arch/riscv/kernel/setup.c     |   4 +-
 arch/riscv/kernel/smpboot.c   |   2 -
 drivers/acpi/numa/srat.c      |   3 +-
 include/linux/acpi.h          |   4 +
 8 files changed, 156 insertions(+), 11 deletions(-)
 create mode 100644 arch/riscv/kernel/acpi_numa.c

arch/riscv/include/asm/acpi.h +13 -2

index 7dad0cf9d701..e0a1f84404f3 100644
@@ -61,11 +61,14 @@ static inline void arch_fix_phys_package_id(int num, u32 slot) { }
 
 void acpi_init_rintc_map(void);
 struct acpi_madt_rintc *acpi_cpu_get_madt_rintc(int cpu);
-u32 get_acpi_id_for_cpu(int cpu);
+static inline u32 get_acpi_id_for_cpu(int cpu)
+{
+     return acpi_cpu_get_madt_rintc(cpu)->uid;
+}
+
 int acpi_get_riscv_isa(struct acpi_table_header *table,
                     unsigned int cpu, const char **isa);
 
-static inline int acpi_numa_get_nid(unsigned int cpu) { return NUMA_NO_NODE; }
 void acpi_get_cbo_block_size(struct acpi_table_header *table, u32 *cbom_size,
                           u32 *cboz_size, u32 *cbop_size);
 #else

@@ -87,4 +90,12 @@ static inline void acpi_get_cbo_block_size(struct acpi_table_header *table,
 
 #endif /* CONFIG_ACPI */
 
+#ifdef CONFIG_ACPI_NUMA
+int acpi_numa_get_nid(unsigned int cpu);
+void acpi_map_cpus_to_nodes(void);
+#else
+static inline int acpi_numa_get_nid(unsigned int cpu) { return NUMA_NO_NODE; }
+static inline void acpi_map_cpus_to_nodes(void) { }
+#endif /* CONFIG_ACPI_NUMA */
+
 #endif /*_ASM_ACPI_H*/

arch/riscv/kernel/Makefile +1 -0

index f71910718053..5d3e9cf89b76 100644
@@ -105,3 +105,4 @@ obj-$(CONFIG_COMPAT)              += compat_vdso/
 
 obj-$(CONFIG_64BIT)          += pi/
 obj-$(CONFIG_ACPI)           += acpi.o
+obj-$(CONFIG_ACPI_NUMA)      += acpi_numa.o

arch/riscv/kernel/acpi.c +0 -5

index e619edc8b0cc..040bdbfea2b4 100644
@@ -191,11 +191,6 @@ struct acpi_madt_rintc *acpi_cpu_get_madt_rintc(int cpu)
      return &cpu_madt_rintc[cpu];
 }
 
-u32 get_acpi_id_for_cpu(int cpu)
-{
-     return acpi_cpu_get_madt_rintc(cpu)->uid;
-}
-
 /*
  * __acpi_map_table() will be called before paging_init(), so early_ioremap()
  * or early_memremap() should be called here to for ACPI table mapping.

arch/riscv/kernel/acpi_numa.c new file +44 -0

new file mode 100644
index 000000000000..493642a61457
@@ -0,0 +1,133 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * ACPI 6.6 based NUMA setup for RISCV
+ * Lots of code was borrowed from arch/arm64/kernel/acpi_numa.c
+ *
+ * Copyright 2004 Andi Kleen, SuSE Labs.
+ * Copyright (C) 2013-2016, Linaro Ltd.
+ *           Author: Hanjun Guo <[email protected]>
+ * Copyright (C) 2024 Intel Corporation.
+ *
+ * Reads the ACPI SRAT table to figure out what memory belongs to which CPUs.
+ *
+ * Called from acpi_numa_init while reading the SRAT and SLIT tables.
+ * Assumes all memory regions belonging to a single proximity domain
+ * are in one chunk. Holes between them will be included in the node.
+ */
+
+#define pr_fmt(fmt) "ACPI: NUMA: " fmt
+
+#include <linux/acpi.h>
+#include <linux/bitmap.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/memblock.h>
+#include <linux/mmzone.h>
+#include <linux/module.h>
+#include <linux/topology.h>
+
+#include <asm/numa.h>
+
+static int acpi_early_node_map[NR_CPUS] __initdata = { NUMA_NO_NODE };
+
+int __init acpi_numa_get_nid(unsigned int cpu)
+{
+     return acpi_early_node_map[cpu];
+}
+
+static inline int get_cpu_for_acpi_id(u32 uid)
+{
+     int cpu;
+
+     for (cpu = 0; cpu < nr_cpu_ids; cpu++)
+             if (uid == get_acpi_id_for_cpu(cpu))
+                     return cpu;

Quoted 9 lines

arch/riscv/kernel/acpi_numa.c line 50

Quoted 6 lines

arch/riscv/kernel/acpi_numa.c new file +6 -0

+
+     return -EINVAL;
+}
+
+static int __init acpi_parse_rintc_pxm(union acpi_subtable_headers *header,
+                                   const unsigned long end)

Please check alignment.

Sure.

Quoted 43 lines

arch/riscv/kernel/acpi_numa.c new file +43 -0

+{
+     struct acpi_srat_rintc_affinity *pa;
+     int cpu, pxm, node;
+
+     if (srat_disabled())
+             return -EINVAL;
+
+     pa = (struct acpi_srat_rintc_affinity *)header;
+     if (!pa)
+             return -EINVAL;
+
+     if (!(pa->flags & ACPI_SRAT_RINTC_ENABLED))
+             return 0;
+
+     pxm = pa->proximity_domain;
+     node = pxm_to_node(pxm);
+
+     /*
+      * If we can't map the UID to a logical cpu this
+      * means that the UID is not part of possible cpus
+      * so we do not need a NUMA mapping for it, skip
+      * the SRAT entry and keep parsing.
+      */
+     cpu = get_cpu_for_acpi_id(pa->acpi_processor_uid);
+     if (cpu < 0)
+             return 0;
+
+     acpi_early_node_map[cpu] = node;
+     pr_info("SRAT: PXM %d -> HARTID 0x%lx -> Node %d\n", pxm,
+             cpuid_to_hartid_map(cpu), node);
+
+     return 0;
+}
+
+void __init acpi_map_cpus_to_nodes(void)
+{
+     int i;
+
+     /*
+      * In ACPI, SMP and CPU NUMA information is provided in separate
+      * static tables, namely the MADT and the SRAT.
+      *
+      * Thus, it is simpler to first create the cpu logical map through

Quoted 9 lines

arch/riscv/kernel/acpi_numa.c line 100

Quoted 7 lines

arch/riscv/kernel/acpi_numa.c new file +7 -0

+      * an MADT walk and then map the logical cpus to their node ids
+      * as separate steps.
+      */
+     acpi_table_parse_entries(ACPI_SIG_SRAT, sizeof(struct acpi_table_srat),
+                                         ACPI_SRAT_TYPE_RINTC_AFFINITY,
+                                         acpi_parse_rintc_pxm, 0);
+

Alignment here as well.

Sure.

Quoted 43 lines

Quoted 5 lines

arch/riscv/kernel/acpi_numa.c new file +5 -0

+     for (i = 0; i < nr_cpu_ids; i++)
+             early_map_cpu_to_node(i, acpi_numa_get_nid(i));
+}
+
+/* Callback for Proximity Domain -> logical node ID mapping */

arch/riscv/kernel/acpi_numa.c line 115

Quoted 10 lines

arch/riscv/kernel/acpi_numa.c new file +10 -0

+void __init acpi_numa_rintc_affinity_init(struct acpi_srat_rintc_affinity *pa)
+{
+     int pxm, node;
+
+     if (srat_disabled())
+             return;
+
+     if (pa->header.length < sizeof(struct acpi_srat_rintc_affinity)) {
+             pr_err("SRAT: Invalid SRAT header length: %d\n",
+                     pa->header.length);

Can we merge these into single line?

Quoted 26 lines

arch/riscv/kernel/acpi_numa.c new file +18 -0

+             bad_srat();
+             return;
+     }
+
+     if (!(pa->flags & ACPI_SRAT_RINTC_ENABLED))
+             return;
+
+     pxm = pa->proximity_domain;
+     node = acpi_map_pxm_to_node(pxm);
+
+     if (node == NUMA_NO_NODE) {
+             pr_err("SRAT: Too many proximity domains %d\n", pxm);
+             bad_srat();
+             return;
+     }
+
+     node_set(node, numa_nodes_parsed);
+}

arch/riscv/kernel/setup.c +0 -0

index 4f73c0ae44b2..a2cde65b69e9 100644
@@ -281,8 +281,10 @@ void __init setup_arch(char **cmdline_p)
      setup_smp();
 #endif

Quoted 6 lines

arch/riscv/kernel/setup.c void __init setup_arch(char **cmdline_p) · line 286
Quoted 4 lines
arch/riscv/kernel/setup.c +2 -1
-     if (!acpi_disabled)
+     if (!acpi_disabled) {
              acpi_init_rintc_map();
+             acpi_map_cpus_to_nodes();
Is it not possible to fill up both in single parsing of MADT?

I think it's not possible to fill both in a single MADT parse since the NUMA info is provided in a separate SRAT table.

For RISC-V, currently we parsed 2 times of the MADT.
FIrst one in setup_smp()->acpi_parse_and_init_cpus() call to build up cpuid_to_hartid_map.
Second one in acpi_init_rintc_map call to build the cpu_madt_rintc[] cache structure.

Since the first one depends on the CONFIG_SMP, I am not sure whether it's possible to combine
these two parts into one.

Quoted 32 lines

arch/riscv/kernel/setup.c +1 -0

+     }
 
      riscv_init_cbo_blocksizes();
      riscv_fill_hwcap();

arch/riscv/kernel/smpboot.c +0 -2

index 519b6bd946e5..b188d83d1ec4 100644
@@ -101,7 +101,6 @@ static int __init acpi_parse_rintc(union acpi_subtable_headers *header, const un
      if (hart == cpuid_to_hartid_map(0)) {
              BUG_ON(found_boot_cpu);
              found_boot_cpu = true;
-             early_map_cpu_to_node(0, acpi_numa_get_nid(cpu_count));
              return 0;
      }

@@ -111,7 +110,6 @@ static int __init acpi_parse_rintc(union acpi_subtable_headers *header, const un
      }
 
      cpuid_to_hartid_map(cpu_count) = hart;
-     early_map_cpu_to_node(cpu_count, acpi_numa_get_nid(cpu_count));
      cpu_count++;
 
      return 0;

drivers/acpi/numa/srat.c +0 -0

index 503abcf6125d..1f0462cef47c 100644
@@ -219,7 +219,8 @@ int __init srat_disabled(void)
      return acpi_numa < 0;
 }

Quoted 7 lines

drivers/acpi/numa/srat.c int __init srat_disabled(void) · line 223
Quoted 3 lines
drivers/acpi/numa/srat.c +2 -1
-#if defined(CONFIG_X86) || defined(CONFIG_ARM64) || defined(CONFIG_LOONGARCH)
+#if defined(CONFIG_X86) || defined(CONFIG_ARM64) || defined(CONFIG_LOONGARCH) || \
+     defined(CONFIG_RISCV)
Please check alignment. Or make it single line if fits in 100 chars. Also, it looks it covers most of the architectures now. Is it possible to simplify / remove the condition? I hope IA64 is removed now?

Good catch!

Since IA64 support was removed in commit cf8e8658100d4(https://lwn.net/Articles/923376/).
I think it's possible to remove the condition. Will fix it in v2.

Thanks,
Haibo

Quoted 28 lines

May be you need to update the comment at #endif too.

Thanks
Sunil

Quoted 23 lines

drivers/acpi/numa/srat.c +0 -0

 /*
  * Callback for SLIT parsing.  pxm_to_node() returns NUMA_NO_NODE for
  * I/O localities since SRAT does not list them.  I/O localities are

include/linux/acpi.h +4 -1

index a65273db55c6..be78a9d28927 100644
@@ -269,8 +269,12 @@ acpi_numa_gicc_affinity_init(struct acpi_srat_gicc_affinity *pa) { }
 
 int acpi_numa_memory_affinity_init (struct acpi_srat_mem_affinity *ma);
 
+#ifdef CONFIG_RISCV
+void acpi_numa_rintc_affinity_init(struct acpi_srat_rintc_affinity *pa);
+#else
 static inline void
 acpi_numa_rintc_affinity_init(struct acpi_srat_rintc_affinity *pa) { }
+#endif
 
 #ifndef PHYS_CPUID_INVALID
 typedef u32 phys_cpuid_t;
--

2.34.1

Sunil V L

2024-03-05 10:06 UTC #10 lore raw

Re: [PATCH 3/4] ACPI: RISCV: Add NUMA support based on SRAT and SLIT

Quoted 247 lines · Haibo Xu

Quoted 129 lines · Sunil V L

Quoted 127 lines · Haibo Xu

Signed-off-by Haibo Xu <[email protected]>

---
 arch/riscv/include/asm/acpi.h |  15 +++-
 arch/riscv/kernel/Makefile    |   1 +
 arch/riscv/kernel/acpi.c      |   5 --
 arch/riscv/kernel/acpi_numa.c | 133 ++++++++++++++++++++++++++++++++++
 arch/riscv/kernel/setup.c     |   4 +-
 arch/riscv/kernel/smpboot.c   |   2 -
 drivers/acpi/numa/srat.c      |   3 +-
 include/linux/acpi.h          |   4 +
 8 files changed, 156 insertions(+), 11 deletions(-)
 create mode 100644 arch/riscv/kernel/acpi_numa.c

arch/riscv/include/asm/acpi.h +13 -2

index 7dad0cf9d701..e0a1f84404f3 100644
@@ -61,11 +61,14 @@ static inline void arch_fix_phys_package_id(int num, u32 slot) { }
 
 void acpi_init_rintc_map(void);
 struct acpi_madt_rintc *acpi_cpu_get_madt_rintc(int cpu);
-u32 get_acpi_id_for_cpu(int cpu);
+static inline u32 get_acpi_id_for_cpu(int cpu)
+{
+     return acpi_cpu_get_madt_rintc(cpu)->uid;
+}
+
 int acpi_get_riscv_isa(struct acpi_table_header *table,
                     unsigned int cpu, const char **isa);
 
-static inline int acpi_numa_get_nid(unsigned int cpu) { return NUMA_NO_NODE; }
 void acpi_get_cbo_block_size(struct acpi_table_header *table, u32 *cbom_size,
                           u32 *cboz_size, u32 *cbop_size);
 #else

@@ -87,4 +90,12 @@ static inline void acpi_get_cbo_block_size(struct acpi_table_header *table,
 
 #endif /* CONFIG_ACPI */
 
+#ifdef CONFIG_ACPI_NUMA
+int acpi_numa_get_nid(unsigned int cpu);
+void acpi_map_cpus_to_nodes(void);
+#else
+static inline int acpi_numa_get_nid(unsigned int cpu) { return NUMA_NO_NODE; }
+static inline void acpi_map_cpus_to_nodes(void) { }
+#endif /* CONFIG_ACPI_NUMA */
+
 #endif /*_ASM_ACPI_H*/

arch/riscv/kernel/Makefile +1 -0

index f71910718053..5d3e9cf89b76 100644
@@ -105,3 +105,4 @@ obj-$(CONFIG_COMPAT)              += compat_vdso/
 
 obj-$(CONFIG_64BIT)          += pi/
 obj-$(CONFIG_ACPI)           += acpi.o
+obj-$(CONFIG_ACPI_NUMA)      += acpi_numa.o

arch/riscv/kernel/acpi.c +0 -5

index e619edc8b0cc..040bdbfea2b4 100644
@@ -191,11 +191,6 @@ struct acpi_madt_rintc *acpi_cpu_get_madt_rintc(int cpu)
      return &cpu_madt_rintc[cpu];
 }
 
-u32 get_acpi_id_for_cpu(int cpu)
-{
-     return acpi_cpu_get_madt_rintc(cpu)->uid;
-}
-
 /*
  * __acpi_map_table() will be called before paging_init(), so early_ioremap()
  * or early_memremap() should be called here to for ACPI table mapping.

arch/riscv/kernel/acpi_numa.c new file +44 -0

new file mode 100644
index 000000000000..493642a61457
@@ -0,0 +1,133 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * ACPI 6.6 based NUMA setup for RISCV
+ * Lots of code was borrowed from arch/arm64/kernel/acpi_numa.c
+ *
+ * Copyright 2004 Andi Kleen, SuSE Labs.
+ * Copyright (C) 2013-2016, Linaro Ltd.
+ *           Author: Hanjun Guo <[email protected]>
+ * Copyright (C) 2024 Intel Corporation.
+ *
+ * Reads the ACPI SRAT table to figure out what memory belongs to which CPUs.
+ *
+ * Called from acpi_numa_init while reading the SRAT and SLIT tables.
+ * Assumes all memory regions belonging to a single proximity domain
+ * are in one chunk. Holes between them will be included in the node.
+ */
+
+#define pr_fmt(fmt) "ACPI: NUMA: " fmt
+
+#include <linux/acpi.h>
+#include <linux/bitmap.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/memblock.h>
+#include <linux/mmzone.h>
+#include <linux/module.h>
+#include <linux/topology.h>
+
+#include <asm/numa.h>
+
+static int acpi_early_node_map[NR_CPUS] __initdata = { NUMA_NO_NODE };
+
+int __init acpi_numa_get_nid(unsigned int cpu)
+{
+     return acpi_early_node_map[cpu];
+}
+
+static inline int get_cpu_for_acpi_id(u32 uid)
+{
+     int cpu;
+
+     for (cpu = 0; cpu < nr_cpu_ids; cpu++)
+             if (uid == get_acpi_id_for_cpu(cpu))
+                     return cpu;

Quoted 9 lines

arch/riscv/kernel/acpi_numa.c line 50

Quoted 6 lines

arch/riscv/kernel/acpi_numa.c new file +6 -0

+
+     return -EINVAL;
+}
+
+static int __init acpi_parse_rintc_pxm(union acpi_subtable_headers *header,
+                                   const unsigned long end)

Please check alignment.

Sure.

Quoted 43 lines

arch/riscv/kernel/acpi_numa.c new file +43 -0

+{
+     struct acpi_srat_rintc_affinity *pa;
+     int cpu, pxm, node;
+
+     if (srat_disabled())
+             return -EINVAL;
+
+     pa = (struct acpi_srat_rintc_affinity *)header;
+     if (!pa)
+             return -EINVAL;
+
+     if (!(pa->flags & ACPI_SRAT_RINTC_ENABLED))
+             return 0;
+
+     pxm = pa->proximity_domain;
+     node = pxm_to_node(pxm);
+
+     /*
+      * If we can't map the UID to a logical cpu this
+      * means that the UID is not part of possible cpus
+      * so we do not need a NUMA mapping for it, skip
+      * the SRAT entry and keep parsing.
+      */
+     cpu = get_cpu_for_acpi_id(pa->acpi_processor_uid);
+     if (cpu < 0)
+             return 0;
+
+     acpi_early_node_map[cpu] = node;
+     pr_info("SRAT: PXM %d -> HARTID 0x%lx -> Node %d\n", pxm,
+             cpuid_to_hartid_map(cpu), node);
+
+     return 0;
+}
+
+void __init acpi_map_cpus_to_nodes(void)
+{
+     int i;
+
+     /*
+      * In ACPI, SMP and CPU NUMA information is provided in separate
+      * static tables, namely the MADT and the SRAT.
+      *
+      * Thus, it is simpler to first create the cpu logical map through

Quoted 9 lines

arch/riscv/kernel/acpi_numa.c line 100

Quoted 7 lines

arch/riscv/kernel/acpi_numa.c new file +7 -0

+      * an MADT walk and then map the logical cpus to their node ids
+      * as separate steps.
+      */
+     acpi_table_parse_entries(ACPI_SIG_SRAT, sizeof(struct acpi_table_srat),
+                                         ACPI_SRAT_TYPE_RINTC_AFFINITY,
+                                         acpi_parse_rintc_pxm, 0);
+

Alignment here as well.

Sure.

Quoted 43 lines

Quoted 5 lines

arch/riscv/kernel/acpi_numa.c new file +5 -0

+     for (i = 0; i < nr_cpu_ids; i++)
+             early_map_cpu_to_node(i, acpi_numa_get_nid(i));
+}
+
+/* Callback for Proximity Domain -> logical node ID mapping */

arch/riscv/kernel/acpi_numa.c line 115

Quoted 10 lines

arch/riscv/kernel/acpi_numa.c new file +10 -0

+void __init acpi_numa_rintc_affinity_init(struct acpi_srat_rintc_affinity *pa)
+{
+     int pxm, node;
+
+     if (srat_disabled())
+             return;
+
+     if (pa->header.length < sizeof(struct acpi_srat_rintc_affinity)) {
+             pr_err("SRAT: Invalid SRAT header length: %d\n",
+                     pa->header.length);

Can we merge these into single line?

Quoted 26 lines

arch/riscv/kernel/acpi_numa.c new file +18 -0

+             bad_srat();
+             return;
+     }
+
+     if (!(pa->flags & ACPI_SRAT_RINTC_ENABLED))
+             return;
+
+     pxm = pa->proximity_domain;
+     node = acpi_map_pxm_to_node(pxm);
+
+     if (node == NUMA_NO_NODE) {
+             pr_err("SRAT: Too many proximity domains %d\n", pxm);
+             bad_srat();
+             return;
+     }
+
+     node_set(node, numa_nodes_parsed);
+}

arch/riscv/kernel/setup.c +0 -0

index 4f73c0ae44b2..a2cde65b69e9 100644
@@ -281,8 +281,10 @@ void __init setup_arch(char **cmdline_p)
      setup_smp();
 #endif

Quoted 6 lines

arch/riscv/kernel/setup.c void __init setup_arch(char **cmdline_p) · line 286
Quoted 4 lines
arch/riscv/kernel/setup.c +2 -1
-     if (!acpi_disabled)
+     if (!acpi_disabled) {
              acpi_init_rintc_map();
+             acpi_map_cpus_to_nodes();
Is it not possible to fill up both in single parsing of MADT?

Quoted 3 lines

I think it's not possible to fill both in a single MADT parse since the NUMA info is provided in a separate SRAT table.

Ahh, yes. My mistake. This looks good to me.

Thanks,
Sunil

Haibo Xu

author

2024-01-31 02:19 UTC #11 lore raw

PATCH 4/4 ACPI: RISCV: Enable ACPI based NUMA

In reply to Haibo Xu

Enable ACPI based NUMA for RISCV in Kconfig.

Signed-off-by Haibo Xu <[email protected]>

---
 arch/riscv/Kconfig        | 1 +
 drivers/acpi/numa/Kconfig | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)

arch/riscv/Kconfig +1 -0

index bffbd869a068..e586ab959f34 100644
@@ -438,6 +438,7 @@ config NUMA
 	select HAVE_SETUP_PER_CPU_AREA
 	select NEED_PER_CPU_EMBED_FIRST_CHUNK
 	select NEED_PER_CPU_PAGE_FIRST_CHUNK
+	select ACPI_NUMA if ACPI
 	select OF_NUMA
 	select USE_PERCPU_NUMA_NODE_ID
 	help

drivers/acpi/numa/Kconfig +1 -1

index 849c2bd820b9..525297c44250 100644
@@ -2,7 +2,7 @@
 config ACPI_NUMA
 	bool "NUMA support"
 	depends on NUMA
-	depends on (X86 || ARM64 || LOONGARCH)
+	depends on (X86 || ARM64 || LOONGARCH || RISCV)
 	default y if ARM64
 
 config ACPI_HMAT

Signature · 1 line

2.34.1

Arnd Bergmann

2024-01-31 09:33 UTC #12 lore raw

Re: [PATCH 4/4] ACPI: RISCV: Enable ACPI based NUMA

drivers/acpi/numa/Kconfig line 5

Quoted 10 lines · Haibo Xu

drivers/acpi/numa/Kconfig +1 -1

index 849c2bd820b9..525297c44250 100644
@@ -2,7 +2,7 @@
 config ACPI_NUMA
 	bool "NUMA support"
 	depends on NUMA
-	depends on (X86 || ARM64 || LOONGARCH)
+	depends on (X86 || ARM64 || LOONGARCH || RISCV)

The dependency is no longer needed now since these are the four architectures that support ACPI now that IA64 is gone.

All of them also 'select ACPI_NUMA' by default, though on x86 this can still be disabled by manually turning off CONFIG_X86_64_ACPI_NUMA. I suspect we don't actually ever want to turn it off on x86 either, so I guess the Kconfig option can just be removed entirely.

Arnd

Haibo Xu

2024-02-01 02:59 UTC #13 lore raw

Re: [PATCH 4/4] ACPI: RISCV: Enable ACPI based NUMA

Quoted 17 lines · Arnd Bergmann

drivers/acpi/numa/Kconfig line 5
Quoted 10 lines · Haibo Xu
drivers/acpi/numa/Kconfig +1 -1
index 849c2bd820b9..525297c44250 100644
@@ -2,7 +2,7 @@
 config ACPI_NUMA
      bool "NUMA support"
      depends on NUMA
-     depends on (X86 || ARM64 || LOONGARCH)
+     depends on (X86 || ARM64 || LOONGARCH || RISCV)
The dependency is no longer needed now since these are the four architectures that support ACPI now that IA64 is gone.

Quoted 6 lines

All of them also 'select ACPI_NUMA' by default, though on x86 this can still be disabled by manually turning off CONFIG_X86_64_ACPI_NUMA. I suspect we don't actually ever want to turn it off on x86 either, so I guess the Kconfig option can just be removed entirely.

Good catch! Will revert the change in the next version.

To remove the dependency entirely, I think we need a separate patch for it. What's your opinion?

Regards,
Haibo

Quoted 1 line

Arnd

Arnd Bergmann

2024-02-01 05:52 UTC #14 lore raw

Re: [PATCH 4/4] ACPI: RISCV: Enable ACPI based NUMA

Quoted 25 lines · Haibo Xu

Quoted 17 lines · Arnd Bergmann
drivers/acpi/numa/Kconfig line 5
Quoted 10 lines · Haibo Xu
drivers/acpi/numa/Kconfig +1 -1
index 849c2bd820b9..525297c44250 100644
@@ -2,7 +2,7 @@
 config ACPI_NUMA
      bool "NUMA support"
      depends on NUMA
-     depends on (X86 || ARM64 || LOONGARCH)
+     depends on (X86 || ARM64 || LOONGARCH || RISCV)
The dependency is no longer needed now since these are the four architectures that support ACPI now that IA64 is gone.
Quoted 6 lines
All of them also 'select ACPI_NUMA' by default, though on x86 this can still be disabled by manually turning off CONFIG_X86_64_ACPI_NUMA. I suspect we don't actually ever want to turn it off on x86 either, so I guess the Kconfig option can just be removed entirely.

Quoted 4 lines

Good catch! Will revert the change in the next version.
To remove the dependency entirely, I think we need a separate patch for it. What's your opinion?

Agreed, that last bit would need to be acked by the x86 maintainers then, as it only affects them in practice.

Arnd

Sunil V L

2024-03-05 05:26 UTC #15 lore raw

Re: [PATCH 4/4] ACPI: RISCV: Enable ACPI based NUMA

In reply to Haibo Xu

Quoted 25 lines · Haibo Xu

Enable ACPI based NUMA for RISCV in Kconfig.

Signed-off-by Haibo Xu <[email protected]>

---
 arch/riscv/Kconfig        | 1 +
 drivers/acpi/numa/Kconfig | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)

arch/riscv/Kconfig +1 -0

index bffbd869a068..e586ab959f34 100644
@@ -438,6 +438,7 @@ config NUMA
 	select HAVE_SETUP_PER_CPU_AREA
 	select NEED_PER_CPU_EMBED_FIRST_CHUNK
 	select NEED_PER_CPU_PAGE_FIRST_CHUNK
+	select ACPI_NUMA if ACPI
 	select OF_NUMA
 	select USE_PERCPU_NUMA_NODE_ID
 	help

drivers/acpi/numa/Kconfig +0 -0

index 849c2bd820b9..525297c44250 100644

drivers/acpi/numa/Kconfig line 5

Quoted 5 lines

drivers/acpi/numa/Kconfig +1 -1

@@ -2,7 +2,7 @@
 config ACPI_NUMA
 	bool "NUMA support"
 	depends on NUMA
-	depends on (X86 || ARM64 || LOONGARCH)
+	depends on (X86 || ARM64 || LOONGARCH || RISCV)

Is it possible to remove this if IA64 is removed now?

Thanks,
Sunil

Quoted 6 lines

drivers/acpi/numa/Kconfig +0 -0
 	default y if ARM64
 
 config ACPI_HMAT
Signature · 1 line
2.34.1

Haibo Xu

2024-03-05 08:32 UTC #16 lore raw

Re: [PATCH 4/4] ACPI: RISCV: Enable ACPI based NUMA

Quoted 25 lines · Sunil V L

Quoted 23 lines · Haibo Xu

Enable ACPI based NUMA for RISCV in Kconfig.

Signed-off-by Haibo Xu <[email protected]>

---
 arch/riscv/Kconfig        | 1 +
 drivers/acpi/numa/Kconfig | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)

arch/riscv/Kconfig +1 -0

index bffbd869a068..e586ab959f34 100644
@@ -438,6 +438,7 @@ config NUMA
      select HAVE_SETUP_PER_CPU_AREA
      select NEED_PER_CPU_EMBED_FIRST_CHUNK
      select NEED_PER_CPU_PAGE_FIRST_CHUNK
+     select ACPI_NUMA if ACPI
      select OF_NUMA
      select USE_PERCPU_NUMA_NODE_ID
      help

drivers/acpi/numa/Kconfig +0 -1

index 849c2bd820b9..525297c44250 100644
--- a/drivers/acpi/numa/Kconfig

Quoted 9 lines

drivers/acpi/numa/Kconfig line 5
Quoted 7 lines
+++ b/drivers/acpi/numa/Kconfig
drivers/acpi/numa/Kconfig +1 -1
@@ -2,7 +2,7 @@
 config ACPI_NUMA
      bool "NUMA support"
      depends on NUMA
-     depends on (X86 || ARM64 || LOONGARCH)
+     depends on (X86 || ARM64 || LOONGARCH || RISCV)
Is it possible to remove this if IA64 is removed now?

Yes. Arnd also suggest removing this totally.
Will update it in v2.

Quoted 8 lines

Thanks,
Sunil
Quoted 6 lines
drivers/acpi/numa/Kconfig +0 -1
      default y if ARM64
 
 config ACPI_HMAT
--
2.34.1

Haibo Xu

2024-03-05 02:30 UTC #17 lore raw

In reply to Haibo Xu

Hi Sunil,

Could you help review this patch set?

Thanks,
Haibo

Quoted 82 lines · Haibo Xu

This patch series enable RISC-V ACPI NUMA support which was based on the recently approved ACPI ECR[1].
Patch 1/4 is generated from the acpica PR[2] and should be merged through the acpica project. Due to this dependency, other 3 patches can only be merged after the corresponding ACPICA patch was pulled into linux.
Patch 2/4 add the common SRAT RINTC affinity structure handler. Patch 3/4 add RISC-V specific acpi_numa.c file to parse NUMA information from SRAT and SLIT ACPI tables.
Patch 4/4 add corresponding ACPI_NUMA config for RISC-V Kconfig.
Based-on: https://github.com/linux-riscv/linux-riscv/tree/for-next
[1] https://mantis.uefi.org/mantis/view.php?id=2433
[2] https://github.com/acpica/acpica/pull/926
Testing:
Since the ACPI AIA/PLIC support patch set is still under upstream review, hence it is tested using the poll based HVC SBI console and RAM disk.
1) Build latest Qemu with the following patch backported
https://lore.kernel.org/all/[email protected]/
https://github.com/vlsunil/qemu/commit/42bd4eeefd5d4410a68f02d54fee406d8a1269b0
2) Build latest EDK-II
https://github.com/tianocore/edk2/blob/master/OvmfPkg/RiscVVirt/README.md
3) Build Linux with the following configs enabled
CONFIG_RISCV_SBI_V01=y
CONFIG_SERIAL_EARLYCON_RISCV_SBI=y
CONFIG_HVC_RISCV_SBI=y
4) Build buildroot rootfs.cpio
5) Launch the Qemu machine
   qemu-system-riscv64 -nographic \
   -machine virt,pflash0=pflash0,pflash1=pflash1 -smp 4 -m 8G \
   -blockdev node-name=pflash0,driver=file,read-only=on,filename=RISCV_VIRT_CODE.fd \
   -blockdev node-name=pflash1,driver=file,filename=RISCV_VIRT_VARS.fd \
   -object memory-backend-ram,size=4G,id=m0 \
   -object memory-backend-ram,size=4G,id=m1 \
   -numa node,memdev=m0,cpus=0-1,nodeid=0 \
   -numa node,memdev=m1,cpus=2-3,nodeid=1 \
   -numa dist,src=0,dst=1,val=30 \
   -kernel linux/arch/riscv/boot/Image \
   -initrd buildroot/output/images/rootfs.cpio \
   -append "root=/dev/ram ro console=hvc0 earlycon=sbi"
[    0.000000] ACPI: SRAT: Node 0 PXM 0 [mem 0x80000000-0x17fffffff]
[    0.000000] ACPI: SRAT: Node 1 PXM 1 [mem 0x180000000-0x27fffffff]
[    0.000000] NUMA: NODE_DATA [mem 0x17fe3bc40-0x17fe3cfff]
[    0.000000] NUMA: NODE_DATA [mem 0x27fff4c40-0x27fff5fff]
...
[    0.000000] ACPI: NUMA: SRAT: PXM 0 -> HARTID 0x0 -> Node 0
[    0.000000] ACPI: NUMA: SRAT: PXM 0 -> HARTID 0x1 -> Node 0
[    0.000000] ACPI: NUMA: SRAT: PXM 1 -> HARTID 0x2 -> Node 1
[    0.000000] ACPI: NUMA: SRAT: PXM 1 -> HARTID 0x3 -> Node 1
Haibo Xu (4):
ACPICA: SRAT: Add RISC-V RINTC affinity structure
ACPI: NUMA: Add handler for SRAT RINTC affinity structure
ACPI: RISCV: Add NUMA support based on SRAT and SLIT
ACPI: RISCV: Enable ACPI based NUMA
 arch/riscv/Kconfig            |   1 +
 arch/riscv/include/asm/acpi.h |  15 +++-
 arch/riscv/kernel/Makefile    |   1 +
 arch/riscv/kernel/acpi.c      |   5 --
 arch/riscv/kernel/acpi_numa.c | 133 ++++++++++++++++++++++++++++++++++
 arch/riscv/kernel/setup.c     |   4 +-
 arch/riscv/kernel/smpboot.c   |   2 -
 drivers/acpi/numa/Kconfig     |   2 +-
 drivers/acpi/numa/srat.c      |  35 ++++++++-
 include/acpi/actbl3.h         |  18 ++++-
 include/linux/acpi.h          |   7 ++
 11 files changed, 209 insertions(+), 14 deletions(-)
 create mode 100644 arch/riscv/kernel/acpi_numa.c
--
2.34.1

Sunil V L

2024-03-05 04:44 UTC #18 lore raw

In reply to Haibo Xu

Hi Haibo,

Quoted 7 lines · Haibo Xu

This patch series enable RISC-V ACPI NUMA support which was based on the recently approved ACPI ECR[1].
Patch 1/4 is generated from the acpica PR[2] and should be merged through the acpica project. Due to this dependency, other 3 patches can only be merged after the corresponding ACPICA patch was pulled into linux.

Quoted 8 lines

Patch 2/4 add the common SRAT RINTC affinity structure handler. Patch 3/4 add RISC-V specific acpi_numa.c file to parse NUMA information from SRAT and SLIT ACPI tables.
Patch 4/4 add corresponding ACPI_NUMA config for RISC-V Kconfig.
Based-on: https://github.com/linux-riscv/linux-riscv/tree/for-next
[1] https://mantis.uefi.org/mantis/view.php?id=2433

Please avoid providing mantis link. It is not useful for people who are not UEFI members. Better to provide the link to the PDF version of the ECR approved.

Thanks,
Sunil

Haibo Xu

2024-03-05 07:42 UTC #19 lore raw

Quoted 15 lines · Sunil V L

Hi Haibo,
Quoted 12 lines · Haibo Xu
This patch series enable RISC-V ACPI NUMA support which was based on the recently approved ACPI ECR[1].
Patch 1/4 is generated from the acpica PR[2] and should be merged through the acpica project. Due to this dependency, other 3 patches can only be merged after the corresponding ACPICA patch was pulled into linux.
Patch 2/4 add the common SRAT RINTC affinity structure handler. Patch 3/4 add RISC-V specific acpi_numa.c file to parse NUMA information from SRAT and SLIT ACPI tables.
Patch 4/4 add corresponding ACPI_NUMA config for RISC-V Kconfig.

Quoted 8 lines

Quoted 3 lines
Based-on: https://github.com/linux-riscv/linux-riscv/tree/for-next
[1] https://mantis.uefi.org/mantis/view.php?id=2433
Please avoid providing mantis link. It is not useful for people who are not UEFI members. Better to provide the link to the PDF version of the ECR approved.

Sure. Will update with a PDF link in v2.

Thanks,
Haibo

Quoted 2 lines

Thanks,
Sunil